Fix DB2 LUW reflection, binary binds and XML/DECFLOAT results

[aminghadersohi]

Problems (found against a local DB2 LUW 11.5.9.0 server)

Reflection (DB2Reflector, LUW only):

• Foreign keys leak across schemas. get_foreign_keys filters the constrained table by name only, so for A.CHILD it also returns the foreign keys of B.CHILD. Columns of a composite key are not paired by key position.
• Column names split. get_pk_constraint and get_indexes split SYSCAT.INDEXES.COLNAMES on \w+, so a column "AMT$X" is reflected as two columns amt and x. DESC index columns lose their ordering. The primary key reports its backing index name (SQL2609...) instead of the constraint name.
• Phantom indexes. DB2's internal XML path index on an XML column is reflected as a user index.
• Unique constraints are not ordered by key position.
• Inspector.has_sequence raises TypeError: has_sequence() got an unexpected keyword argument 'info_cache'.
• get_check_constraints is not implemented.
• Types. DECFLOAT, BINARY and VARBINARY reflect as NullType with a warning; CHAR/VARCHAR FOR BIT DATA reflect as CHAR/VARCHAR although the DBAPI returns bytes.

ibm_db dialect:

• Binary executemany corrupts data. SQLAlchemy binds binary values through dbapi.Binary, which ibm_db_dbi implements as memoryview. In executemany, ibm_db rejects a memoryview for BINARY/VARBINARY/FOR BIT DATA (SQL0302N) and stores its repr text (<memory at 0x...>) in BLOB columns. Single-row execute is fine. Reproduced with ibm_db 3.2.3 and 3.3.0. Binding bytes works in both paths.
• DECFLOAT results are str; they are now Decimal (or float with asdecimal=False), and Decimal binds are sent as exact text.
• XML results start with a byte order mark and a <?xml version="1.0" encoding="UTF-16" ?> declaration synthesized by the CLI (even for a document stored with a UTF-8 declaration). DB2 stores XML without a declaration, so exactly that prefix is dropped.
• Disconnects. A connection lost while fetching surfaces as the base ibm_db_dbi.Error, which is_disconnect ignored, so the dead connection went back to the pool. After a server restart, close() on a dead connection raises CLI0106E, and the pool logged an error for every connection it discarded; that one error is now ignored in do_close.

IBM i (AS400Reflector) and z/OS (OS390Reflector) are unchanged. The default-schema issue is handled separately in #203.

Tests

New test/test_luw_reflection.py, same style as the existing tests: connectionless processor/keyword checks and live reflection and round-trip checks (fixtures in two schemas with same-named tables, a composite FK with reversed key order, a "AMT$X" column, a DESC index, an XML column, a check constraint, a sequence, DECFLOAT/BINARY/VARBINARY/FOR BIT DATA/BLOB values via executemany).

Same test file, Python, SQLAlchemy, DBAPI and server; only ibm_db_sa/ differs between upstream master 7f3866e and this patch:

	master	this patch
test/test_luw_reflection.py	12 FAIL / 1 PASS	13 PASS

(The unique-constraint test passes on master for this fixture; master already joins TABCONST/KEYCOLUSE on schema and table.)

python -m pytest -p sqlalchemy.testing.plugin.pytestplugin --dburi "db2+ibm_db://user:pass@host:port/db" -q test/test_luw_reflection.py

Validation: Python 3.11, SQLAlchemy 2.0.52, ibm-db 3.2.3, pytest 9.1.1, local throwaway DB2 Community Edition 11.5.9.0 container. The full legacy suite was not run.